JPH10254489A - Speech recognition system for numerals - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide the speech recognition system which prevents the transmission of the numerals that are incorrectly recognized. SOLUTION: The system is provided with a control device 33 which recognizes at least one of numeral columns and generates the recognized numerals of at least one of numeral columns. If there exists at least one erroneously recognized numeral of a first numeral column, the device 33 compares a second numeral column against the first numeral column. If the number of numerals in the second numeral column is less than the number of numerals in the first numeral column, the device 33 determines the related numerals of the portion of the first numeral column which has the numeral that best matches with the numeral of the second numeral column and the device 33 replaces the numerals of the determined portion of the first numeral column by the unmatched numerals of the second numeral column.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a speech recognition system for numbers, comprising a control device for recognizing at least one digit sequence and generating a recognized digit of said at least one digit sequence.

[0002]

2. Description of the Related Art Such a system is disclosed, for example, in WO 9
5/06309 A1 and comprises a remote control including a microphone and a circuit for forming a modulated infrared signal. The voice input of the user captured by the microphone is transmitted to the control device via an infrared signal,
The control device converts the speech input into a code word,
It is transmitted to an evaluation circuit which forms control commands for, for example, a video cassette recorder or a television receiver. Certain functions can be performed on the television receiver or video cassette recorder by individual voice inputs or voice commands. For example, a channel can be selected, a volume level can be set, or video tape playback can be stopped. A description is also given of the time programming of the videocassette recorder, wherein the programming for inputting the channel, date, start and end times is performed in a predefined, sequential order. When inputting voice, it is necessary to input numbers. For example, after a number has been entered for a channel or time, a response is provided giving individual control of the video cassette recorder or television receiver.
Next, comparison with the stored pattern is performed. If the numbers are misrecognized and assigned to the wrong pattern, there is incomplete control.

[0003]

SUMMARY OF THE INVENTION It is therefore an object of the present invention to provide a speech recognition apparatus which avoids the transmission of incorrectly recognized digits.

[0004]

The object of the present invention is to provide a method, comprising: at least one misrecognized first digit sequence number;
The controller compares the dictated second number string with the first number string, and when the number of numbers in the second number string is less than the number of numbers in the first number string, the controller is Determining the associated number of the portion of the first number sequence that has the most number that matches the number of the second number sequence;
This is achieved by a system of the type defined in the introduction, wherein the mismatched digits of the second sequence are replaced by the digits of the determined part of the first sequence.

In the system according to the present invention, the user performs verification of the voice input. Make selective corrections to these numbers that are not recognized. The speech recognition is described in the publication "Herm
annNey, Volker Steinbiss, Xavier Aubert, Reinhold
Haeb-Umbach: Progress in Large Vocabulary, Continuo
us Speech Recognition, in: H. Niemann, R. de Mori,
G. Hanrieder: Progress and Prospects of Speech Re
search and Technology, 1994, pp. 75 to 92 ". According to this method, a sequence of linked numbers is recognized with the help of a hidden Markov model. After generating the sequence of entered digits for verification, the user can accept or reject the recognized digit sequence, and then re-enter the particular digit. The number is generated by speech synthesis by the controller or by a previously entered and stored number. The controller shall understand the numbers 0 to 9 and certain control inputs such as, for example, "Yes", "No" and the like.

When recognizing the first sequence of digits, the user is asked if the sequence is understood correctly. If not, request the user to provide another audio input. At this time, the user may input a completely new number string or only a partial number string. Thereafter, the first number string is compared with the newly input second number string. Next, the controller determines the portion of the first number sequence that has a number that matches the most number of the second number sequence. At this time, it is a necessary condition that the number of numbers in the second number string is smaller than the number of numbers in the first number string. Thereafter, the mismatched digits in the second digit sequence are substituted for the digits in the first digit sequence.

[0007] Such a system forms, for example, a telephone section in which telephone numbers are formed by voice input. Furthermore, the system according to the invention may be used in enhanced services (eg language selection in a network).

[0008] The system according to the invention has the advantage that the user enters such a number with an individual context only in the case of a correction. For example, the user pronounces only the digits before and after the misunderstood number. This form of correction corresponds to the natural behavior that the user is accustomed to, and is faster than re-entering the entire digit string. In addition, this form of correction is more likely to be successful because the entry of partial digit strings reduces the risk of identification errors.

During the evaluation process, the control device determines the number of digits in the first and second digit sequences, and determines which digits of all relevant parts of the first digit sequence are of the second digit sequence. Determine if it matches the number. If various parts or sub-strings of the first number sequence have the same number of matches, one of these sub-columns is selected for correction. The first sub-column may be selected from a plurality of sub-columns having the same number of numbers matching the second number string.

[0010] Further, the control device is used to mark at least one number of the first number string to be replaced with the number of the second number string, and to pronounce the marked number. Other numbers are also pronounced with certain stress. The controller indicates the numbers having odd positions in the sequence of numbers with increasing stress and the numbers having even positions and the last position of the numbers with decreasing stress. This natural form of pronunciation with paired prosody can improve the recording of numbers. Pronunciation with corrected numerical emphasis (constant stress) can make successful proof easier.

By generating the corrected first digit sequence after the evaluation of the first and second digit sequence, the controller forms a question for the user as to whether the digit sequence has been correctly recognized.

[0012] The invention also relates to a method for recognizing a digit, which recognizes at least one digit sequence and generates a recognized digit of the at least one character sequence. At this time,
In an event where at least one digit of the first digit sequence is incorrectly recognized, the dictated second digit sequence is compared with the first digit sequence, and the number of digits in the second digit sequence is changed to the first digit sequence. If the number of numbers is less than the number of the second character string, determine the correction of the number of the part of the first character string that has the most number of numbers that match the number of the second number string, It is provided to substitute for the digits of the determined part of the first digit sequence.

[0013] These and other aspects of the invention will be apparent from and elucidated with reference to the embodiments described hereinafter.

[0014]

FIG. 1 shows a preferred embodiment of a digitized speech recognition system, which comprises a microphone 1, two amplifiers 2 and 3, and a speech recognition device 4.
, An evaluation circuit 5 and a loudspeaker 6. The speech recognition device 4 and the evaluation circuit 5 form a control device 33. The voice input of the user is supplied to the microphone 1. The system includes a particular speech input, ie, a particular sequence of digits (eg, “3 8 7 4 21 6”) and a control input supplied to the speech recognizer 4 via the amplifier 2. The speech recognizer 4 may, for example, comprise a signal processor with suitable peripherals, whose execution program enables speech recognition. Such programs are known and the methods on which they are based are described, for example, in the literature "Hermann Ney, Volker Steinb
iss, Xavier Aubert, Reinhold Haeb-Umbach: Progress
in Large Vocabulary, Continuous Speech Recognitio
n, in: H. Niemann, R. de Mori, G. Hanrieder: Progr
ess and Prospects of Speech Research and Technolog
y, 1994, pp. 75 to 92 ". Recognize the digit string input by the user and input it to the evaluation circuit 5 as a code word (for example, in an ASCII code). The evaluation circuit 5 includes a voice response unit that forms a voice response from the recognized digits. The voice response unit may be a voice synthesizer for inputting the synthesized numbers to the amplifier 3, and the voice synthesis unit retrieves the stored voice syllables of the speaker from the memory and sends these voice syllables to the amplifier 3. You may enter it.

The voice response is notified to the user via the loudspeaker 6, and these are tested. The evaluation circuit also generates certain notices or complaints, such as, for example, "Is these numbers exactly understood?" At this time, the user may execute a correction in an event in which a number or a string of numbers is misunderstood.

The evaluation circuit 5 further includes a microprocessor having suitable peripherals, the microprocessor comprising a software module for processing the recognized control inputs and digits and for controlling the voice response unit. FIG. 2 shows a flowchart of the main processing for recognizing numbers. The digit sequence is recognized by the speech recognition device 4 and is identified in block 7 of FIG. ZN
, Then parsing and accent marking (AN AK, block 8) is performed for the voice response unit. The number having an even position in the number sequence is marked with "b" and the number having an odd position in the number sequence is marked with "e". The last digit of the sequence is marked with an "e", regardless of whether it has an even or odd position in the sequence. This means that the numbers at the first, third, fifth, etc. positions of the number sequence are characterized by "b" and the numbers at the second, fourth, sixth position, etc. of the number sequence are characterized by "e". At this time, a pair-style prosody is generated for the response. For example, the numeric string “3 8 7 4 2 1
6 "to" 3b 8e 7b 4e 2b 1e 6
e ".

In the flowchart shown in FIG. 2, the next step after block 8 is called block 9 (AU KO)
Indicated by This block is characterized by a response of the recognized number sequence and a question as to whether the number sequence was correctly recognized. When responding to the digit sequence, the voice response unit of the evaluation circuit 5 uses two phrase variants. The numbers occur on ascending or descending stress. For words marked "b", use a complaint with ascending stress,
For the character marked "e", the phrase is used in descending stress. As a result, there is a paired prosodic pattern in the voice response unit, which pattern corresponds to the natural way of speaking of humans.

Once the user has answered the question about what the system is aware of (block 10, E
RK A), question block 11 (OK?) Tests what the answer is. If the user replies “yes”, the number string is recognized and the input is terminated. Next, further processing may be performed using the recognized number sequence. If the answer is "No", the system requests a correction from the user,
Abbreviation AU in 2 Expressed by FR. At this time, the user may input a completely new number string or a number sub-string. The subsequent speech recognition process and analysis of the newly entered digit sequence is referred to in block 13 by the abbreviation ER
K Indicated by ZK. After speech recognition and speech analysis in block 13, other analysis and stress marking as described above are performed (block 8).

The analysis represented by block 13 will be further described with the aid of the flow charts of FIGS. The start of the analysis in FIG. 3 is shown as ST. First, the length L (Z1) of the previous number string Z1 is changed to the new number string Z.
Test whether it is shorter than the length L (Z2) of block 2 (block 14: L (Z1) <L (Z2)). If not, the new digit sequence Z2 replaces the previous digit sequence Z1, which is indicated in block 15 by Z1 → Z2. this is,
The analysis (EN) ends. However, if the sequence Z1 is longer than or equal to the sequence Z2, the variables m, mT and mS are set to zero as shown in block 16 (m = 0, mT = 0, mS = 0). .

Here, a description will be given of a part of the flowchart for finding which part of the previous number string Z1 is most common with the new number string Z2. At the beginning of the first loop, it is tested whether the sequence Z2 has been compared with each part of the sequence Z1 (block 17). In block 17, it is checked whether the value of the variable m is less than or equal to the difference between the lengths of the digit strings Z1 and Z2: m ≦ L (Z1) −L (Z
2). For example, if the number string Z1 is the number “3 8 7 4 2
If the number string Z2 has the number "752", the length of the number string Z1 is equal to 7 and the length of the number string Z2 is equal to 3. Therefore, the five parts of the numeric string Z1 (“387”, “874”, “742”, “421”)
And the first loop should be passed a total of five times to compare & 216 with the digit sequence Z2. Block 17
If the comparison in indicates that the value of the variable m is greater than the difference between the lengths of the numeric strings Z1 and Z2, the first loop is ended and the processing is switched to the second loop whose flowchart is shown in FIG. The change to the second loop is indicated by the mark "A".
(Block 18).

If the comparison m≤L (Z1) -L (Z2) is true, set two other variables n and t to zero, as shown in block 19. The variable n indicates the position of the number in the number string Z2, and the variable t indicates the number of numbers that match the part of the number string Z1 to be compared with the number string Z2. The next question block 20 consists of blocks 21, 22 and 23
Is a part of the sub-loop as well. In block 20, it is checked whether the value of the variable n is smaller than the length of the digit string Z2. If so, the interrogation block 21 determines that the number at position m + n of the number string Z1 is equal to the number at position n of the number string Z2 (Z1 (m + n) = Z2
(N)). If the question is answered affirmatively, the variable t is incremented (block 22). If the answer is no, the process jumps to block 23 as if block 22 had been processed. Block 23
Indicates the increment of variable n. Thereafter, processing is performed in block 20.

If the value of the variable n is equal to or greater than the length of the numeric string Z2 (block 20), processing is performed in a question block 24. Here, it is tested whether the value of the variable t is larger than the value of the variable mT. If so, the variable mT is set equal to t and the variable mS is set equal to m (block 25). The variable mS indicates a portion having the largest number of numbers that match the number string Z2 of the number string Z1. The variable mT is equal to the number of matching digits. In the next step, after the negative result of the question in block 24, or
After setting the variables mT and mS at, the variable m is incremented at block. This ends the first loop for determining the part of the number string Z1 most corresponding to the number string Z2. In the example specified above, the number "3 8
The part having the number “7 42” in the number string Z 1 having “7 42 16” corresponds most to the number string Z 2 having the number “75 2”.

The second loop shown in the flowchart of FIG. 4 marks a part of the numeral string Z1 which is different from the numeral Z2. The flowchart shown in FIG. 4 starts with mark “A” in block 27. Prior to the start of this second loop, the variable n is set to zero, which is indicated in block 28. This variable n indicates the position of the numeral in the numeral string Z2. The second loop comprises question blocks 29 and 30, and other blocks 31 and 32. In a question block 29, it is asked whether the value of the variable n is smaller than the length of the digit string Z2 (n <L (Z2)).
If this is not the case, the analysis ends. If so, it tests whether the number in said part of the number sequence Z1 is equal to the assigned number of the number sequence Z2 (block 30). The mathematical expression for this is Z1 (m + n) =
Z2 (n). If the number at position m + n of the number string Z1 corresponds to the number at position n of the number string Z2, block 32 is processed. In other cases, if the numbers do not correspond, the number at position n of the number string Z2 is substituted for the number at position n + m of the number string Z1. In this case, in block 31, the expression Z1 (m +
n) → Z2 (n). In addition, reference numbers "a" are appended to the replaced numbers. This relationship is shown in block 31 by the expression aZ1 (m + n). In the next step, the variable n is incremented as shown in block 32. Next, further processing is performed again in question block 29.

After analysis in block 13 (FIG. 2), a new digit sequence Z1 is supplied to block 8. For example, a new numeric string Z1 having the number “3 8 7 5 2 1 6” is replaced with the number “3 8 7 4 2 1 6”
And the number “75” in the number string Z1 before
2 ". At this time, the number “5” is replaced with the number “4”. In addition, block 8 receives from block 13 the marking of said alternative number or the number indicated by the letter "a". At block 8, these numbers are marked with the letters "b" and "e" as described above. An audio response corresponding to the marked digit sequence is generated at block 9. The number marked "b" is pronounced ascending and the number marked "e" is pronounced as descending. The number marked "a" is further emphasized to indicate the change to the user. For example, the marking of the new numeral string Z1 is described as “3b 8e 7b a5
e 2b 1e 6e ".

As shown in the flowchart of FIG.
Through 13 until the user approves the result.

It should further be appreciated that the control device 33 may be arranged as a computer system performing the functions of the speech recognition unit 4 and the evaluation circuit 5.

[Brief description of the drawings]

FIG. 1 is a block diagram of a speech recognition system.

FIG. 2 is a flowchart illustrating a speech recognition system.

FIG. 3 is a flowchart illustrating a speech recognition system.

FIG. 4 is a flowchart illustrating a speech recognition system.

[Explanation of symbols]

 Reference Signs List 1 microphone 2, 3 amplifier 4 voice recognition device 5 evaluation circuit 6 loudspeaker 33 control device

 ──────────────────────────────────────────────────続 き Continued on the front page (72) Inventor Nils Renke Germany 53332 Bornheim Rathausstraße 8 (72) Inventor Jörg Oker Germany 52066 Aachen Bentstrasse 27 a

Claims (6)

[Claims] 1. A number speech recognition system, comprising: a controller for recognizing at least one digit sequence and generating a recognized digit of said at least one digit sequence.
If there is at least one misrecognized number in the number sequence, the controller compares the dictated second number sequence with the first number sequence and determines that the number of digits in the second number sequence is the first number. If less than the number of digits in the sequence, the controller determines the associated number of the portion of the first sequence having the sequence that best matches the number in the second sequence; Wherein the mismatched digits of the second digit sequence are substituted for the digits of the determined portion of the first digit sequence. 2. The system according to claim 1, wherein the controller comprises:-determining the number of digits in the first and second digit sequences;-all relevant parts of the first digit sequence. Of the number
Determining a match with the second digit string; and-if more parts of the first digit string have the same number of matching digits, the first with the most matches.
A system for selecting a portion of a number sequence. 3. The system according to claim 1, wherein the control device marks at least one alternative number in the first sequence of numbers with a number of the second sequence of digits and sounds the marked number with a sound. A system for use in generating. 4. The system according to claim 1, wherein if a corrected first digit sequence is generated, after evaluating the first and second digit sequences, the control device can recognize the digit sequence correctly. A system for generating a question for a user as to whether or not they are in use. 5. The system of claim 1, wherein the controller generates numbers at odd positions of the sequence of numbers with ascending force and generates numbers at even positions of the sequence of numbers with descending force. A system characterized in that it is used. 6. A speech recognition method for recognizing at least one digit sequence and generating a recognized digit of said at least one digit sequence, wherein at least one digit of the first digit sequence is incorrectly recognized. Comparing the dictated second number string with the first number string, if the number of numbers in the second number string is less than the number of numbers in the first number string,
Determining a correction of a number in the portion of the first string that has the most digits that match the digits of the second string;
The method of claim 1 wherein the mismatched digits of the second digit sequence replace the digits of the determined portion of the first digit sequence.